OpenMausBot is the free Grok Bot alternative everyone's been waiting for, and it just dropped.
By the end of this page you'll have your own team of AI agents doing real work for you.
Reading your emails, doing your research, updating your projects — while you're at the gym.
It's free, and it runs on the AI tools you already use.
You download one app, and you've got a whole team.
There's one feature inside it that changes everything, and most people are going to miss it completely.
Stick with me and you'll walk away with your own AI team, no matter who you are.
500apps you can plug in
3CLIs it runs on
0new subscriptions
2days after Grok Bot
§1 ────── what it is
A whole team of agents, living in a chat app.
It's open source, and it runs on the AI subscriptions you already pay for — your Claude, your Codex, even Grok itself.
No new account. No waitlist. No token limit shutting you down after twenty minutes.
What you're looking at: one contact in the sidebar, opened up. Every bot carries its own five things — a name, a model, its own memory, its own connected apps, and its own computer.
One assistant in one box was version one. This is version two.
§2 ────── the problem
The Twenty Minute Problem.
Grok Bot went into early beta this week.
AI agents that work like teammates — they sign into your tools and come back with finished work.
Sounds great.
Here's the problem. Most people can't get in. There's a waitlist.
And when you do get in, the limits hit fast.
I signed up for the free trial. Within about twenty minutes, it ran out of tokens.
Twenty minutes. You get a taste of what agent teams feel like, and then the door closes.
The Contact List Workforce breaks that cycle, because it never had a meter in the first place.
What you're looking at: the same job, two doors. One runs a meter down to zero and stops. The other runs on the plan you already bought.
Thinking it?"Free usually means worse."
Not here. The agents run on the exact same Claude, Codex and Grok you'd be paying for anyway.
Free is the wrapper, not the brain.
§3 ────── two days later
That gap lasted about two days.
A builder called Milind Soni looked at Grok Bot, built an open source version of the whole idea, and shared it publicly.
Instead of needing a new subscription, it plugs into the AI tools you already use and are already logged into.
What you're looking at: the gap between a paid beta and a free clone of it, measured in days. This is the speed everything moves at now.
§4 ────── the plan
What you're about to get.
What it is, how the team setup works, where it beats Grok Bot and where it falls short.
Then how it slots into a full Agent OS so all your agents share one memory.
The chat-app ideaA model per agentYour Claude skills, inheritedIts own computerAsk-before-it-acts500+ connected appsWired into the Agent OSOne shared brainThe honest limits
§5 ────── the core idea
It looks like WhatsApp. Every contact is an agent.
There's a sidebar of contacts, and every one of them goes and does work, comes back, and shows you what it did.
What you're watching: my real roster on my Mac. Clicking between contacts — each one is a separate agent with its own thread, its own model and its own history.
§6 ────── why this design wins
Your mum can use it. That's the whole trick.
Everyone already knows how to use a chat app, so the technical barrier basically disappears.
New agent? Click the plus sign. New workflow? Type what you want.
Old way~ a weekend
Install a framework, then read its docs
Write a config file for every agent
Wire keys, models and tools by hand
Hit an error message you've never seen
Give up before the first agent runs
Result: you never get a team
New way~ 5 minutes
Open the app you just dragged into Applications
Click the plus sign to add a contact
Name it, pick its model from a dropdown
Type what you want in plain English
Watch it work, tap Allow when it asks
Result: a team, tonight
Thinking it?"I'm not technical enough for this."
If you can message a friend, you can run an agent. That's the entire skill.
The barrier you're picturing came down this week.
§7 ────── a brain per bot
Different jobs need different brains.
With Grok Bot you get the model you're given. Here, every single agent has its own model picker.
Fable 5, Opus, Sonnet, Haiku, Codex or Grok 4.5 — per agent.
What you're watching: the real picker on one bot. The rail down the left switches provider — Claude's models, then Codex's GPT-5.6 line, then Grok — and the tick marks which one this contact runs on.
What you're looking at: the right brain on the right job, instead of forcing one assistant to be good at everything.
§8 ────── a simple example
One agent whose only job is your inbox.
Mine reads my vault every morning and hands me the three things that actually matter, with the first action for each.
What you're watching: my real bots inside the Agent OS. The Inbox Agent's morning brief — three real priorities pulled from my own notes. Then the Build Agent, which wrote a file. Then Vault Scout, which wrote to the shared memory. Real jobs, real output, not an empty chat.
What you're looking at: the morning loop. It reads, it drafts, it waits for your tap. Nothing sends without you.
§9 ────── the part most people miss
It inherits everything you've already built in Claude.
Because it plugs into your existing Claude setup, it pulls in the skills you've already loaded.
I asked my SEO agent for striking-distance keywords. It used my keyword research skill, pulled live Search Console data, and handed me a table. I set up nothing.
What you're watching: a real live Google Search Console pull, running inside the Agent OS. Real queries, real positions, real impressions — then the finished table lands in the workspace where I can open it.
What you're looking at: the thing nobody tells you. Every skill and setup already sitting in your Claude carries over into every agent you create here.
Thinking it?"I already have Claude — why would I need this?"
Inside Claude, your custom workflows are buried in old conversations you have to dig for.
Here your SEO researcher is a name in the sidebar, every day, with everything saved in one place. Same intelligence, completely different usability.
Open the Computer panel and your agent gets a real cloud Linux desktop it drives by itself, with a live preview of the screen.
You can point it at your own Mac instead, or click open desktop and take over yourself.
What you're watching: the Computer panel opening on a real bot. This is where its desktop lives, and where you watch it click around while you do something else.
What you're looking at: the three places an agent's hands can be — a cloud box, your own Mac, or nowhere at all.
§12 ────── honest reality check
Computer use is slow right now.
Grok Bot has the same feature, and when I tested it, it got stuck opening Google and typing hello.
It works, it's real, but you're not handing it your whole day yet.
What you're looking at: where this actually sits. A year ago it barely existed for normal people. This week two apps shipped it wrapped in a chat window. The speed comes later; the shape is already here.
§13 ────── it asks first
Nothing risky happens without your tap.
When an agent wants to run a command or touch a file, a little card pops up in the chat and you tap Allow or Deny.
When I tested it, it asked before touching my Obsidian vault. I allowed it, and off it went.
What you're watching: a real agent stopped mid-job, waiting on me. I tap Allow inside the Agent OS and it carries straight on. Nothing risky moved until I said so.
What you're looking at: every action goes through one gate, and you are the gate.
Thinking it?"Approving things constantly sounds annoying."
It is, a bit. But it's open source, so you can ask Claude itself to edit the program and change how approvals work.
You own the code. That's a level of control Grok Bot will never give you.
§14 ────── connected apps
Over 500 apps, one click each.
Gmail, Slack, Notion, Linear and hundreds more, through a marketplace built on Composio.
Connect an app once, and every agent on your team can use it as a tool.
What you're watching: the real Connected apps drawer on my Mac — Google Docs, Drive, Notion, Linear, Sentry, PostHog, Discord, X, Reddit, each with a Connect button. The orange line is honest: I haven't pasted a Composio key yet, so nothing is connected.
§15 ────── where your stuff lives
The app is local first.
One small server lives on your Mac and runs every agent process, and your conversations, keys and logs sit in a folder on your machine.
To be clear: you're still using cloud models, because that's where Claude, Codex and Grok run. The app itself stays with you.
What you're looking at: the honest split. Everything in the gold box is on your Mac. Only the thinking leaves.
§16 ────── getting it
Download, drag, open. That's the install.
Grab the Mac app from GitHub, drag it into Applications, open it. The server is built in.
The one thing you need is at least one AI tool already installed and signed in — Claude Code, Codex, or Grok.
What you're watching: the real first run on my Mac. It finds the CLIs by itself — Claude Code 2.1.229, Codex 0.144.1, Grok Build 1.0.3, all ticked — and then the roster opens.
# the only line you'd ever type — and only if you'd rather run it from source
git clone https://github.com/milind-soni/OpenMausBot && cd OpenMausBot && pnpm install && pnpm dev:server
Mac only today. Windows and Linux aren't built yet — the server itself is portable Node, so those are the obvious next ports.
An app gives you agents. A system gives you agents that remember.
§17 ────── wired into the agent os
I built it into the Agent OS. One brain, many workers.
On its own, every app is its own island — your OpenMausBot agents don't know what Hermes did last night.
Inside the Agent OS they all sit on one Obsidian vault, so what one agent finds, every other agent can read.
What you're watching: the new OpenMausBot tab inside my Agent OS. Same roster, same bots, same threads as the app — plus a Vault memory switch. I flip it off, flip it back on, hand the bot a job, and it writes its answer straight into the shared memory file.
What you're watching: the Workspace tab. Everything my bots built, in one place — the live Search Console table on top, and underneath, an animated dashboard the Build Agent wrote from one sentence, running live in the page. This is the bit an app on its own can't give you.
What you're looking at: every agent reading from and writing to the same vault. That's the difference between five apps and one system.
Agentic OS/Memories/2026-08-13.md — the real file, after the run
# 2026-08-13 — Agentic OS Memories- **Asked**: "Who is Julian and what is he building right now?" (Vault Scout / OpenMausBot)- **Found**: Answered from `04 Resources/About Julian.md` (canonical profile)- **Next agent note**: read About Julian.md first for any Julian context task
- OpenMausBot demo bot (Vault Scout) checked in.
What you're looking at: the actual lines an OpenMausBot agent wrote into my vault during this build. The green line landed while I was recording the clip above.
Thinking it?"Doesn't running an Agent OS burn a fortune in tokens?"
No — that's the biggest myth about it. It runs the everyday work on a free local model on your own machine, free APIs slot in for more, and for the heavy jobs it drives the CLIs you already pay for. Your Claude subscription already includes the Claude CLI, so you're not paying twice.
It's a layer on top of what you already own, not a new meter. And there are full token-efficiency tutorials inside the Boardroom.
§18 ────── the real lesson
Adding it took about thirty minutes.
The Agent OS uses the same four ingredients for every agent, so a new tool just joins the team.
If your setup is built around one app, you rebuild every time. If it's built around a system, every new tool makes it stronger.
What you're looking at: the four ingredients. Every agent in the OS is wired the same way, which is why a brand-new app slots in in half an hour.
§19 ────── from anywhere
The agents don't stop because you left the house.
The Agent OS runs on my home Mac, and through Tailscale I open the same dashboard from my laptop or my phone.
Same tasks, same chat history, same memory.
What you're looking at: one machine doing the work, and every screen you own looking at it.
§20 ────── the honest limits
Now let's be fair. It's early.
01
Three engines only
Claude, Codex and Grok. The good news: you use the CLI you're already signed into, so no API key. That's a real edge over Hermes, where Claude needs an API.
02
It literally just launched
It's an open source project, so it's messy in places — the formatting of its questions was rough when I tested it. Grok Bot is the polished version. This is the one without the wall.
03
Agents can't tag each other in
I asked one agent to speak to another and it didn't understand. Right now your agents work solo. I'm sure that gets fixed — the code is open and it's the most requested feature in every agent tool.
Tested on my own machine while writing this page. Where it fell short, it says so.
§21 ────── where this leaves you
Three projects, one conclusion.
Grok Bot is built on this shape. OpenMausBot is built on this shape. The Agent OS is built on this shape.
When everyone lands in the same place from different directions, that tells you where things are going.
What you're looking at: version one was one assistant in one box. Version two is a team, each with a name, a job, a model, apps and a computer.
§22 ────── three beliefs to drop
What's actually holding you back.
Wrong: "This stuff changes too fast — whatever I learn is outdated in a month."
Right: The tools change fast, but the shape underneath is settling. Agents with jobs, models, apps, memory. Learn the shape once and every new tool takes you thirty minutes instead of thirty days.
Wrong: "I'll wait until the good one is out of beta."
Right: The people waiting for things to slow down will be learning the basics while everyone else is three tools ahead. Grok Bot dropped, and two days later a free version existed.
Wrong: "I'm not technical enough to run a team of agents."
Right: It's a chat app with AI contacts. If you can message a friend, you can run an agent. That's the whole point of this page.
Don't take my word for it
Members post their wins every day — agency owners, ecom founders, course creators, solo operators across 38 countries. Real businesses, real numbers, in their own words.
The AI Profit Boardroom is where the Agent OS lives — OpenMausBot wired in, shared memory, the 30-day roadmap and live walkthroughs of every big release the week it drops.
